Received 1999 January 27; accepted 1999 September 29

Preprint typeset using LATEX style emulateapj v. 04/03/99

COSMIC CONCORDANCE AND QUINTESSENCE

Limin Wang1 , R. R. Caldwell2 , J. P. Ostriker3 , and Paul J. Steinhardt2

arXiv:astro-ph/9901388 v2 14 Jan 2000

Received 1999 January 27; accepted 1999 September 29

ABSTRACT
We present a comprehensive study of the observational constraints on spatially flat cosmological models
containing a mixture of matter and quintessence a time varying, spatially inhomogeneous component
of the energy density of the universe with negative pressure. Our study also includes the limiting case of
a cosmological constant. We classify the observational constraints by red shift: low red shift constraints
include the Hubble parameter, baryon fraction, cluster abundance, the age of the universe, bulk velocity
and the shape of the mass power spectrum; intermediate red shift constraints are due to probes of
the red shift-luminosity distance based on type 1a supernovae, gravitational lensing, the Lyman-alpha
forest, and the evolution of large scale structure; high red shift constraints are based on measurements
of the cosmic microwave background temperature anisotropy. Mindful of systematic errors, we adopt
a conservative approach in applying these observational constraints. We determine that the range of
quintessence models in which the ratio of the matter density to the critical density is 0.2 <
0.5
m <
and the effective, density-averaged equation-of-state is 1 w <
0.2,
are
consistent
with
the
most

reliable, current low red shift and microwave background observations at the 2 level. Factoring in the
constraint due to the recent measurements of type 1a supernovae, the range for the equation-of-state
is reduced to 1 w <
0.4, where this range represents models consistent with each observational
constraint at the 2 level or better (concordance analysis). A combined maximum likelihood analysis
suggests a smaller range, 1 w <
0.6. We find that the best-fit and best-motivated quintessence
models lie near m 0.33, h 0.65, and spectral index ns = 1, with an effective equation-of-state
w 0.65 for tracker quintessence and w = 1 for creeper quintessence.
problem and prediction of a nearly scale-invariant spectrum of energy density fluctuations. Finding these arguments compelling4 , we will adopt the ansatz that the
universe is spatially flat.
In our previous work (Ostriker & Steinhardt 1995) we
identified the range of models consistent with then-current
observations, restricting attention to the case where E is
vacuum energy or cosmological constant (). The cosmological constant, a static, homogeneous energy component with positive energy density but negative pressure,
was introduced initially by Einstein in a flawed attempt
to model our universe as a static spacetime with positive
spatial curvature. In a spatially flat universe, however, the
negative pressure results in a repulsive gravitational effect
which accelerates the expansion of the universe. This earlier analysis, which preceded by several years the recent
evidence based on luminosity-red shift surveys of type 1a
supernovae, already found that is favored over standard
cold dark matter or open models. Over the intervening
years, the supernovae measurements as well as other observations have reinforced these conclusions.
In this paper, we update our earlier concordance analysis and expand it to include the possibility that the additional energy component consists of quintessence, a dynamical, spatially inhomogeneous form of energy with negative pressure (Caldwell et al. 1998). A common example

1. INTRODUCTION

The most widely studied cosmological models at the

present time are variants of the Cold Dark Matter (CDM)
paradigm within which adiabatic perturbations in a dominant CDM species grow due to gravitational instability
from quantum fluctuations imprinted during an inflationary era. The bulk of the evidence today strongly favors
models within which m < 1 and any hot component is
not significant, HDM CDM . Several authors (Ostriker
& Steinhardt 1995, Krauss & Turner 1995, Turner & White
1997) have found that the best and simplest fit concordant
with current observations is provided by
m = [CDM ] + [baryon ]
[0.30 0.10] + [0.04 0.01]

(1)

with h = 0.65 0.15. One is thus led to either an open

universe or one in which the remaining energy density required to produce a geometrically flat universe is some
additional energy component (E) with
m + E = 1.

(2)

An important advantage of the flat model is that it is consistent with standard inflationary cosmology, and its associated resolution of the cosmological horizon and flatness
1 Department

of Physics, Columbia University, 538 West 120th Street, New York, NY 10027
of Physics, Princeton University, Princeton, NJ 08544
3 Department of Astrophysical Sciences, Princeton University, Princeton, NJ 08544
4 While inflationary cosmologies can be constructed which lead to a spatially open universe (Gott 1982, and e.g. Linde 1999 and references
therein), provided careful tuning, choice of inflaton potential, and/or anthropic arguments, we consider here only the standard case of a spatially
flat universe.
2 Department

WANG ET AL.

is the energy of a slowly evolving scalar field with positive

potential energy, similar to the inflaton field in inflationary
cosmology (Weiss 1987, Ratra & Peebles 1988, Wetterich
1995, Frieman et al. 1995, Coble et al. 1997, Ferreira &
Joyce 1997, Caldwell et al. 1998, Zlatev et al. 1998). Unlike
a cosmological constant, the dynamical field can support
long-wavelength fluctuations which leave an imprint on
the cosmic microwave background (CMB) and large-scale
structure. In particular, the long wavelength fluctuations
change the relation between the amplitude of the CMB
anisotropy and the gravitational potential fluctuations so
that the COBE normalization of the mass power spectrum
depends on the pressure of the quintessence component. A
further distinction is that w, the ratio of pressure (p) to
energy density (), is 1 < w 0 for quintessence whereas
w is precisely 1 for a cosmological constant. Hence, the
expansion history of the universe for a model versus a
quintessence model (for the same m today, say) is different. In general, the acceleration, the age and the volume
of the universe are less for quintessence models than for
models (assuming all other cosmic parameters are fixed).
A prime motivation for considering quintessence models
is to address the coincidence problem, the issue of explaining the initial conditions necessary to yield the nearcoincidence of the densities of matter and the quintessence
component today. For the case of a cosmological constant, the only possible option is to finely tune the ratio
of vacuum density to matter-radiation density to 1 part
in 10120 at the close of inflation in order to have the
correct ratio today. Symmetry arguments from particle
physics are sometimes invoked to explain why the cosmological constant should be zero (Banks 1996) but there is
no known explanation for a positive, observable vacuum
density. For quintessence, because it couples directly to
other forms of energy, one can envisage the possibility of
interactions which may cause the quintessence component
to naturally adjust itself to be comparable to the matter density today. In fact, recent investigations (Zlatev et
al. 1998, Steinhardt et al. 1999) have introduced the notion of tracker field models of quintessence which have
attractor-like solutions (Peebles & Ratra 1988, Ratra &
Peebles 1988) which produce the current quintessence energy density without the fine-tuning of initial conditions.
A related development has been creeper field models
(Huey & Steinhardt 1999) which are nearly as insensitive
to initial conditions but indistinguishable from today.
Fundamental physics provides some further motivation
for light scalar fields. Particle physics theories with dynamical symmetry breaking or non-perturbative effects
have been found which generate potentials with ultralight masses which support negative pressure (Affleck et
al. 1985, Hill & Ross 1988a, Hill & Ross 1998b, Binetruy
et al. 1996, Barreiro et al. 1998, Binetruy 1999). These
suggestive results lend appeal to a particle physics basis
for quintessence, as a logical alternative to an ad hoc invocation of a cosmological constant. We do not aim to base
our investigation of the properties of quintessence cosmologies on a specific particle physics model, however, as such
models are still in a developmental stage. An intriguing
thought is that progress in the cosmological observations
and experiments discussed here will soon decide the issue, possibly pointing to new fundamental physics inaccessible in the accelerator laboratory. We would empha-

Vol. 530

size that scalar field models of quintessence are not only

the simplest, well-motivated choice from a particle physics
standpoint, but, also, they can mimic fluids with arbitrary
equation-of-state. (It was shown explicitly in Caldwell et
al. 1998 that there is a one-to-one correspondence between
a general time-dependent equation-of-state and an equivalent scalar field potential.) Generalizations to tensor fields
or more general stress tensor (Hu 1999) or topological defects (Spergel & Pen 1996) have also been considered; for
the purposes of this study based on current observations,
they can be well described by scalar fields, in addition.
Extending the realm of cosmological models to include quintessence opens up a new degree of freedom,
the equation-of-state of the missing energy component.
The added degree of freedom necessarily makes the procedure of selecting viable models more complicated. Previous studies have considered some specific combinations
of observations (Silveira & Waga 1997, Turner & White
1997, Garnavich et al. 1998, Perlmutter et al. 1998) or
some specific models (Ferreira & Joyce 1997, Ferreira &
Joyce 1998). Here we systematically examine the most
complete range yet of measures of astrophysical phenomena at low, intermediate, and high red shift and make a
complete search in parameter space to objectively identify
the viable models. We show that the observations are consistent with and quintessence models for a substantial
range of parameters. An impressive feature is how a number of observations, not only the measurements of type 1a
supernovae, favor a missing energy component with substantially negative pressure.
The pace of cosmological observations is proceeding so
rapidly that any quantitative conclusions may soon become dated. Nevertheless, we think that an assessment at
the present time is worthwhile for at least three reasons.
First, our study shows a concordance among a growing
number of observations. Compared to the previous analysis (Ostriker & Steinhardt 1995) new constraints have
been added and old ones have been revised, and, yet, notably, the key conclusions a new energy component and
an accelerating expansion rate have been significantly
strengthened. Second, the study isolates those observations which are playing the lead roles in shaping the current conclusions and identifies observations or combinations thereof which will be most decisive in the near future. Third, this comprehensive analysis enables one to
identify specific best-fit models which can be explored in
much greater detail to search for more subtle implications
and tests.
The organization of the paper is as follows. In section 2,
we discuss the parameterization of the cosmological constant and quintessence cosmological models. In section 3,
we present the observational constraints, classified by low,
intermediate, and high red shift. We evaluate the constraints, presenting the results in section 4. We conclude
in section 5 with an identification of the overall best-fit
models and a discussion of future observations.
2. PARAMETERIZATION OF QUINTESSENCE
COSMOLOGICAL MODELS

The quintessence (QCDM) cosmological scenario is a

spatially-flat FRW space-time dominated by radiation at
early times, and cold dark matter (CDM) and quintessence
(Q) at late times. For simplicity, we will consider models in

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

which the quintessence component consists of a scalar field

slowly rolling down its effective potential with a constant
equation-of-state. The detailed equations-of-motion are
discussed in Caldwell et al. 1998. This class of models is a
good approximation for most of the range of quintessence
candidates. The models we consider can then be fully
characterized by the following five parameters:
wQ : A constant quintessence equation-of-state, in the
range w [1, 0] in the present epoch. In most cases, the
quintessence equation-of-state changes slowly with time,
but the observational predictions are well approximated
by treating w as a constant, equal to
Z
Z
w
e da Q (a) w(a)/ da Q (a).
(3)

We will comment on exceptions later. Tracker field

quintessence models have a lower bound on the range of
w, while creeper quintessence leads to w very close to 1,
which is effectively indistinguishable from a true cosmological constant.
m : The matter density parameter, defined as the ratio
of the matter energy density, including CDM and baryons,
to the critical energy density c = 3H 2 /8G, where H is
the Hubble constant. We assume that any contribution to
the energy density due to a hot component, such as neutrinos, is small, insofar as free-streaming has a negligible
effect on the clustering of CDM. Unless otherwise specified,
we have imposed m + Q = 1, where Q is the corresponding density parameter for quintessence. Hence, the
matter density parameter lies in the range m [b , 1].
b : The baryon density parameter, defined as the ratio of the baryonic energy density to the critical energy
density.
h: The Hubble parameter, related to the Hubble constant by H = 100 h km/s/Mpc.
ns : The index of the power spectrum of primordial density fluctuations in the matter and radiation. This parameter also controls the contribution of tensor perturbations,
for which we consider two cases. In the first case, we impose the inflationary relation between the amplitude of the
primordial density and gravitational wave (tensor) perturbations for ns 1, revised for the case of quintessence
(Caldwell & Steinhardt 1998). For ns > 1, we assume the
tensor contribution is negligible. In the second case, we
assume the tensor contribution is negligible for all values
of the spectral index. (We only illustrate the first case; the
second case yields an indistinguishable result in regard to
our concordance analysis.)
The most revealing way to depict the concordance of
constraints on quintessence models is to project the fivedimensional parameter space into the m -w plane. In displaying this plane, we assume the universe is spatially flat.
Since the flatness condition requires Q = 1 m , the parameters w and m completely specify the quintessence
portion of the cosmology. CDM corresponds to the line
w = 1, and SCDM corresponds to the line m = 1 in
this plane.
3. OBSERVATIONAL CONSTRAINTS

We take a conservative approach in applying the cosmological constraints. Observational cosmology is currently
in a period of rapid growth so that the current constraints

must be considered as work in progress, rather than final.

Certainly, measurements of many astrophysical phenomena are becoming more refined, with greater precision as
statistical errors are reduced, and with greater accuracy
as systematic errors are better understood. If this situation described all observations, we could confidently apply
the results to the full limits of the published errors. Because the observational constraints typically restrict combinations of independent model parameters, then by combining several experimental results, we could find tighter
parameter bounds than if the observations were applied
individually. For example, combining two constraints restricts a two-parameter system to an ellipse in parameter
space, whereas the individual constraints applied successively allow a rectangle which contains that ellipse. The
former combination would allow the determination of a
model which was best in the maximum likelihood sense.
However, not all measurements have well controlled systematic errors or even high precision statistical errors. Nor
do they have true gaussian errors. Nor are the errors uncorrelated. A prime example is the current set of type
1a supernovae magnitude-red shift data, which must be
interpreted with caution. For these reasons, we do not advocate cosmological parameter extraction with the current
set of astrophysical data. Until the coming generation of
precision experiments are on-line, we believe it can be misleading to combine the full set of astrophysical data as if
the errors were statistical and gaussian.
For this reason, we employ an additional procedure,
which we call concordance, to evaluate the observational
constraints. We identify models as passing the observational constraints if they lie within 2 of each individual
constraint. (We do not consider the joint probabilities
spanning two or more observations.) We allow a generous
range for systematic errors. Not only does this procedure
provide a reliable picture of current constraints, but there
is the added advantage that it renders transparent which
observations are most important in delimiting the range
of currently allowed parameters and which future observations will be most influential.
One might characterize the difference between the two
approaches as follows: two identical and observationally
independent constraints on a single parameter do not
change, at all, the allowed range of that parameter in the
concordance analysis, but do reduce the range in a maximum likelihood approach. Clearly, the concordance approach is too conservative if the errors are known to be
gaussian. However, in the current case, systematics dominate and the correlations between errors in different observations is unclear. In these circumstances, as discussed in
the Appendix, the maximum likelihood approach can produce seriously misleading results. One should be especially
watchful and examine closely situations where the concordance and maximum likelihood analyses strongly disagree.
Maximum likelihood has the undesireable feature that it
will seek a compromise among conflicting data. Hence,
we advocate the more conservative concordance approach,
supplemented with comparison to a full maximum likelihood estimator for the constraints assuming gaussian errors.
We classify the observations by red shift. At low red
shift, z 1, the constraints are due to: the Hubble constant; age of the universe; baryon density; x-ray cluster

WANG ET AL.

abundance; shape of the mass power spectrum; and bulk

flow. At intermediate red shift, z 1, the constraints are
due to: type 1a supernovae; evolution of the x-ray cluster
abundance; gravitational lensing; Lyman- forest. At high
red shift, z 1, the constraints are due to: the fluctuation
amplitude and spectral tilt based on the large angle CMB
temperature anisotropy (COBE); small angle CMB temperature anisotropy. The high red shift CMB constraint
due to COBE, along with the set of low red shift results,
serve as the strongest, most reliable constraints. These
constraints will be shown to dominate the boundary of
the allowed parameter range. The intermediate red shift
constraints, due to SNe and the evolution of x-ray clusters, are rapidly reaching the point where they impact the
range of cosmological models. The small angle CMB measurements, as well, are soon to yield prime cosmological
information. We will consider each of these constraints in
turn.
The observational constraints used to restrict the
quintessence parameter space are listed in the following
subsection. The low red shift, and COBE-based high red
shift constraints compose the core set of concordance tests
of our cosmological models. The remaining intermediate and high red shift constraints are less certain at the
present, although they offer the promise of powerful discrimination between models in the near future.
3.1. Low Redshift
H: The Hubble constant has been measured through
numerous techniques over the years. Although there has
been a marked increase in the precision of extragalactic
distance measurements, the accurate determination of H
has been slow. The H0 Key Project (Freedman et al. 1998),
which aims to measure the Hubble constant to an accuracy of 10%, currently finds H = 73 6(stat) 8(sys)
km/s/Mpc; the method of type 1a supernovae gives H =
63.1 3.4(internal) 2.9(external) km/s/Mpc (Hamuy et
al. 1996); a recent measurement based on the SunyaevZeldovich effect in four nearby clusters gives H = 54 14
km/s/Mpc (Myers et al. 1997); typical values obtained
from gravitational lens systems are H 5070 km/s/Mpc
with up to 30% errors (Falco et al. 1997, Keeton
& Kochanek 1997, Kundic et al. 1997, Schechter et al.
1997). Using surface brightness fluctuations to calibrate
the bright cluster galaxy Hubble diagram, a far-field value
H = 89 10 km/s/Mpc, out to 11, 000 km/s, has been
obtained (Lauer et al. 1998). Clearly, convergence has not
yet been reached, although some methods are more prone
to systematic uncertainties. Based on these diverse measures, our conservative estimate for the Hubble parameter
is H = 6515 km/s/Mpc at effectively the 2 level. While
knowledge of H would certainly be a decisive constraint,
our current uncertainty will not prevent us from narrowing
the field of viable cosmological models.
t0 : Recent progress in the dating of globular clusters
and the calibration of the cosmic distance ladder has relaxed the lower bound on the age of the universe. We
adopt t0 9.5 Gyr as a 95% lower limit (Chaboyer et al.
1998, Salaris & Weiss 1998), although we note that some
arguments (Paczynski 1999) suggest a 10% higher limit is
more appropriate.
BBN: Recent observations of the deuterium abundance
by Burles and Tytler yield D/H = 3.4 0.3(stat) 105

Vol. 530

(Burles & Tytler 1997a, Burles & Tytler 1997b, Burles &
Tytler 1998). If this value reflects the primordial abundance, then big bang nucleosynthesis (BBN; for a review
see Schramm & Turner 1998 and references therein) with
three light neutrinos gives b h2 = 0.019 0.002, where
the 1 error bars allow for possible systematic uncertainty. While other observations suggest that the abundance varies on galactic length scales where it is expected
to be uniform, suggesting that heretofore unknown processes may be processing the deuterium (Jenkins et al.
1999), we will adopt the hypothesis that the cosmological
abundance has been ascertained by the measurements of
Burles & Tytler.
BF: Observations of the gas in clusters have been used
to estimate the baryon fraction (compared to the total
mass) to be fgas = (0.06 0.003)h3/2 (Evrard 1997;
also see White et al. 1993, Fukugita et al. 1997). The
stellar fraction is estimated to be less than 20% of the
gas fraction, so that fstellar = 0.2h3/2 fgas . Next, simulations suggest that the baryon fraction in clusters is
less than the cosmological value by about 10% (Lubin
et al. 1996) representing a depletion in the abundance of
baryons in clusters by a factor of 0.9 0.1. Hence, the
cosmological baryon fraction (b /m ) is estimated to be
fbaryon = (0.067 0.008)h3/2 + 0.013 at the 1 level. Using the observed baryon density from BBN, we obtain a
constraint on m :
m =



0.019h2
1

0.32
0.067h3/2 + 0.013

(4)

at the 2 level. For h = 0.65, this corresponds to a value

of m = 0.32 0.1.
The baryon fraction has also been estimated on smaller
scales such as galaxy systems (see McGaugh 1998 and references therein). However, the relationship between the local baryon fraction on those scales and the global baryon
fraction is uncertain and currently beyond the power of
numerical study.
8 : The abundance of x-ray clusters at z = 0 provides
a model dependent normalization of the mass power spectrum at the canonical 8h1 Mpc scale. The interpretation
of x-ray cluster data for the case of quintessence models
has been carried out in detail in Wang & Steinhardt 1998,
in which case the constraint is expressed as
8 m = (0.5 0.1) 0.1

(5)

where the error bars are 2, with

= 0.21 0.22w + 0.33m + 0.25
= (ns 1) + (h 0.65).

(6)

This fitting formula is valid for the range of parameters

considered in this paper.
Perhaps the two most important constraints on the mass
power spectrum at this time are the COBE limit on large
scale power and the cluster abundance constraint which
fixes the power on 8h1 Mpc scales. Together, they fix
the spectral index and leave little room to adjust the power
spectrum to satisfy other tests.
Shape: If light traces mass with a constant bias factor
on large scales, then the deprojected APM galaxy cluster

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

data (Peacock 1997b) can be used to constrain the shape

of the underlying mass power spectrum. The bias factor
is defined as b2 PAP M /Pmass , the ratio of the APM to
mass power spectra on a given scale; we assume the bias
represents a constant, quadratic amplification of the clustering power of rare objects over the density field on large
scales (Kaiser 1986). Hence, in keeping with our spirit
of conservativism, we restrict our attention to wavenumbers which are well within the linear regime, using only the
seven lowest frequency bins (as given in Peacock 1997a) for
scales above 8 Mpc/h. The shape constraint consists of the
requirement that the mass power spectrum fit the seven
APM data points with b 1 and a reduced 2 2.0, corresponding to a confidence level of 95%. In effect, our shape
test depends also on the power spectrum amplitude. The
lower bound on the bias is due to the assumption that the
bright, luminous APM objects are preferentially formed
in highly overdense regions (Davis et al. 1985, Bardeen
1986, Kaiser 1986, Cen & Ostriker 1998, Blanton et al.
1998). (Although the bias may be very large at the time of
formation, simple arguments indicate that by the present
time, b may have decreased to no lower than unity. See
Fry 1986, Tegmark & Peebles 1998). The consequences of
b < 1 will be discussed. While we give no upper bound on
b, almost all the best fitting values for concordance models fall within the rough (model-dependent) upper bound
estimate of b <
1.5 based on higher-order statistics of the
APM data set and current theoretical ab initio modeling
(Gaztanaga & Frieman 1994, Cen & Ostriker 1998, Blanton et al. 1998). Our computations show that the popular
shape parameter m h is not an accurate description
of the goodness of fit to the APM data, given the variety
of models that we consider here, since the amplitude of the
power spectrum is characterized by other combinations of
parameters, including w. As illustrated in Figure 1, five
sample models with ranging from 0.20 to 0.52 all pass
our shape test based on a 2 analysis.
Velocity Field: The large-scale velocity field has long
been used as a means to probe the background and fluctuation matter density field. One method is to compare
peculiar velocity data obtained from distance indicators,
such as Tully-Fisher, and from red shift surveys, in order to estimate m , modulo an assumption about biasing.
Through this method, the quantity f (m )/b is obtained, where f d ln m /d ln a 0.6
m and b is a linear
bias parameter. A variety of recent results (Davis, Nusser,
& Willick 1993, Willick et al. 1997, Willick & Strauss
1997, da Costa et al. 1998) find 0.5 0.6. For a bias
not too different from unity, these results suggest a low
matter density. In the lack of a more complete understanding of bias, however, this method cannot be transformed
into a rigorous constraint on m . As well, it remains to
be understood why similar approaches which compare the
density fields obtained from distance indicators and red
shift survey data, along the lines of the POTENT method
(Bertschinger & Dekel 1989), generally obtain higher values, e.g. 0.9 (Sigad et al. 1998). Another method is
to compare observations with the predicted velocity field
within the context of a particular cosmological model. Results based on the Mark III (Zaroubi et al. 1997) and
SFI (Zehavi 1998, Freudling et al. 1999) catalogs yield
the constraint f 2 P (k) = (4.8 1.5) 103 (Mpc/h)3 and
(4.4 1.7) 103 (Mpc/h)3 at k = 0.1 h/Mpc, respec-

tively. Extrapolating to smaller scales, within the class

of CDM models, the SFI constraint can be recast as
8 f = 0.82 0.12. Due to the discrepancy with the cluster
abundance constraint, we are hesitant to apply this recent
result until further analysis reinforces its conclusions.
The bulk flow on the largest scales provides another
method to do cosmology with the velocity field. In Figure 2
we compare the predictions of a set of QCDM models with
observation. The large sample variance on the Maxwellian
distributed velocity field means that consistency requires
the observations lie below the upper 95%CL bulk velocity.
(The lower 95%CL bulk velocity is very small, so we may
effectively treat this constraint as an upper bound.) A
measurement near or above the swath of predicted curves
could serve as a strong indicator of the cosmology. For
comparison, we also show the measured bulk velocities of
Dekel et al. 1999, Giovanelli et al. 1998, and Lauer & Postman 1994. At present, the diversity of measurements, as
displayed in Figure 2, dilutes the strength of the constraint
resulting from the comparison of the velocity dipole with
the CMB dipole.
3.2. Intermediate Redshift
SNe: Type 1a supernovae are not standard candles,
but empirical calibration of the light curve - luminosity relationship suggests that the objects can be used
as distance indicators. There has been much progress
in these observations recently, and there promises to be
more. Hence, a definitive constraint based on these results would be premature. However, we examine the recent results of the High-Z Supernova Search Team (HZS:
Riess et al. 1998, Garnavich et al. 1998) and the Supernova Cosmology Project (SCP: Perlmutter et al. 1998) to
constrain the luminosity distance - red shift relationship in
quintessence cosmological models. We have adopted the
following data analysis procedure: we use the supernova
data for the shape of the luminosity - red shift relationship
only, allowing the calibration, and therefore the Hubble
constant, to float; we excise all SNe at z < 0.02 to avoid
possible systematics due to local voids and overdensities;
for SNe at z > 0.02, we assume a further uncertainty,
added in quadrature, corresponding to a peculiar velocity
of 300km/s in order to devalue nearby SNe relative to the
more distant ones (for the SCP data, a velocity of 300km/s
has already been included). There is substantial scatter in
the supernovae data, as seen in Figure 3. The scatter is
so wide that no model we have tested passes a 2 test
with the full SCP data set; using a reduced set, Fit C,
argued by the SCP as being more reliable Perlmutter et
al. 1998, a finite range of models do pass the 2 test and
the range is comparable to the range obtained by the 2
test using the HZS data set. To gauge the current situation, we will report both 2 tests and maximum likelihood
tests; to be conservative, we use the largest boundary (the
2 test based on HZS data using MLCS analysis) for our
concordance constraint.
Cluster Evolution: The abundance of rich clusters
objects presumed to have formed from high density peaks
drawn from the exponential tail of an initially Gaussian
perturbation distribution can be used to constrain the
amplitude of the mass power spectrum at intermediate red
shift. The current observations have been converted into a
number density of clusters above a certain mass threshold

WANG ET AL.

M1.5 , defined to be the mass within the comoving radius

Rcom = 1.5h1 Mpc. For the models of interest, the abundance evolves approximately as a power law for 0 < z < 1:
n(> M1.5 , z) 10A(M1.5 )z

(7)

(see Wang & Steinhardt 1998). The bigger A(M1.5 ) is, the
weaker the evolution is, implying low m and w. Since
the measurements (summarized in Bahcall et al. 1997)
are still in the preliminary stage, we adopt A(M1.5 =
8 1014 h1 Mpc) = 1.7+1.5
2.1 as a conservative, 2 limit.
Similar tests have been applied in the context of CDM
and open models (Bahcall et al. 1997, Carlberg et al. 1997),
for which the results are model dependent.
Lensing Counts: The statistics of multiply imaged
quasars, lensed by intervening galaxies or clusters, can be
used to determine the luminosity distance - red shift relationship, and thereby constrain quintessence cosmological
models. There exists a long literature of estimates of the
lensing constraint on models (e.g. from Turner et al.
1984 to Falco et al. 1998). In one approach, the cumulative lensing probability for a sample of quasars is used to
estimate the expected number of lenses and distribution
of angular separations. Using the Hubble Space Telescope
Snapshot Survey quasar sample (Maoz et al. 1993), which
found four lenses in 502 sources, Maoz-Rix (Maoz & Rix
1993) arrived at the limit <
0.7 at the 95% CL. In
a series of studies, similar constraints have been obtained
using optical (Kochanek 1995, Kochanek 1996) and radio
lenses (Falco et al. 1998). Waga and collaborators (Torres & Waga 1996, Waga & Miceli 1999) have generalized
these results, finding that the constraint weakens for larger
values of the background equation-of-state, w > 1 (as
noted earlier by Ratra & Quillen 1992). In our evaluation
of the constraint based on the HST-SSS data set, we find
that the 95% confidence level region is approximately de2
scribed by Q <
0.75 + (1 + w) , until the inequality is
saturated at w = 1/2, consistent with Torres & Waga
1996, Waga & Miceli 1999. In principle, this test is a sensitive probe of the cosmology; however, it is susceptible to
a number of systematic errors (for a discussion, see Malhotra et al. 1997, Cheng & Krauss 1999). Uncertainties
in the luminosity function for source and lens, lens evolution, lensing cross section, and dust extinction for optical lenses, threaten to render the constraints compatible
with or even favor a low density universe over m = 1.
Taking the above into consideration, none of the present
constraints on quintessence due to the statistics of multiply imaged quasars are prohibitive: models in concordance
with the low-z constraints are compatible with the lensing
constraints.
Ly- (and other mass power spectrum measurements
at moderate red shift, 1 < z < 10): The Ly- forest
has been used as the basis of a number of cosmological
probes. Most recently, the effect of the local mass density in the intergalactic medium on the Ly- optical depth
(Hui 1999, Croft et al. 1998a) has been used to estimate
the mass power spectrum at a red shift of z = 2.5 (Croft et
al. 1998b, Weinberg et al. 1998). This is a good pedagogical example to study what can and cannot be learned from
studies at moderate red shift 1 < z < 10. In Figure 4 we
show the linear mass power spectrum today (upper panel)
and at red shift z = 2.5 (lower panel) for the representa-

Vol. 530

tive best-fit models discussed in Table I. The mass power

spectra in the upper panel all satisfy COBE normalization at large scales and the cluster abundance constraint
on 8 on 8 h1 Mpc scales. Since the models have already
passed the constraints on 8 h1 Mpc scales and higher,
one might hope that tests of the linear power spectrum at
smaller scales might further distinguish the models. However, smaller scales correspond to the non-linear growth
regime where effects like scale-dependent bias make it difficult to compare observations to the linear power spectrum. Measurements of the Lyman- forest are promising
because they probe the power spectrum on smaller scales
at a red shift before non-linearities develop and, hence, enable direct comparison to the linear power spectrum. We
draw the readers attention to the fact that, in converting
to z = 2.5, we have made a model-dependent rescaling of
the abscissa so that the units are those of velocity, which
then allows direct comparison to the data. In the upper
(z = 0) panel, the models differ substantially on large
scale but appear to converge on small scales. Projecting
back to z = 2.5 and rescaling, one might hope that the
models are distinct due to the differing growth functions.
However, instead, examples with the same spectral tilt
and m H0 /H(z) (as is demonstrated inadvertently by our
representative models) nearly overlap everywhere, making
discrimination very difficult. When m H0 /H(z) is fixed,
as shown in the figure, then, we note, the Lyman- measurements can be used to determine the tilt, ns .
The current data appears to favor ns = 1. There is some
hope that improved limits can discriminate among models with different tilt, but determining other parameters,
and especially discriminating between cosmological constant and quintessence on this basis, appears hopeless because the predictions of various models converge, as shown
in the lower panel. Of course, this conclusion applies not
only to Lyman- forest measurements, but any approach
that measures the mass power spectrum at moderate red
shift.
Further applications of the Ly- forest, such as the
abundance of damped Ly- absorbers (Gardner et al.
1997) and correlations in the lines of sight at red shifts
z 2 4 (McDonald & Miralda-Escude 1998, Hui et al.
1999) have been developed as tests of geometry and expansion history, although no substantial constraints have
as yet been obtained.
There are a number of observational probes which are
sensitive to the cosmology, but which have not yet matured into critical tests. We list some of these tests which
may prove to be powerful constraints in the near future.
Measures of the abundance of objects, similar to the cluster evolution constraint, can be used to gauge the growth
of structure. Observations of galaxies formed as early as
z
> 3 (Steidel et al. 1998) have been interpreted, on the
basis of a Press-Schechter formalism (Press & Schechter
1974), to suggest that among a family of CDM cosmologies, flat, low-density models best satisfy the constraint
(Mo & Fukugita 1996).
Finally, it has been proposed to use the statistics of gravitational lens arcs produced by intermediate red shift clusters as a means to distinguish cosmological models (Wu
& Mao 1996, Bartelmann et al. 1998). These are potentially extremely powerful tests, insofar as the rare, high

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

density fluctuations reflect the underlying cosmology. In

particular, the lens arcs statistics are exponentially sensitive to the growth function, which differs for versus
quintessence models and, hence, has the potential of distinguishing the two scenarios. However, results based on
numerical simulations are not yet capable of resolving the
core of clusters accurately enough to provide reliable limits
(Wambsganss et al. 1998) from the theoretical side.
3.3. High Redshift
One of the most powerful cosmological probes is the
CMB anisotropy, an imprint of the recombination epoch
on the celestial sphere. The large angle temperature
anisotropy pattern recorded by COBE can be used to place
two constraints on cosmological models.
COBE norm: The observed amplitude of the CMB
power spectrum is used to constrain the amplitude of the
underlying density perturbations. We adopt the method
of Bunn & White 1997 to normalize the power spectrum to
COBE. As we use a modified version of CMBFAST (Seljak
& Zaldarriaga 1996) to compute the CMB anisotropy spectra, this normalization is carried out automatically.) We
have verified that this method, originally developed for
and open CDM models, can be applied to the quintessence
cosmological models considered in this work (Dave 1998).
Of course, there is uncertainty associated with the COBE
normalization: the 2 uncertainty in rms quantities is
approximately 20% (see footnote #4 in Bunn & White
1997), which conservatively allows for statistical errors, as
well as the systematic uncertainty associated with the differences in the galactic and ecliptic frame COBE map pixelizations, and potential contamination by high-latitude
foregrounds(e.g. Gorski et al. 1998).
ns : COBE has been found to be consistent with a
ns = 1.2 0.3 spectral index (Gorski et al. 1996, Hinshaw et al. 1996), but this assumes the only large angular scale anisotropy is generated via the Sachs-Wolfe
effect on the last scattering surface. This neglects the
baryon-photon acoustic oscillations, which produce a rise
in the spectrum, slightly tilting the spectrum observed by
COBE. In general, the spectral index determined by fitting
the large angular scale CMB anisotropy of a quintessence
model, which is also modified by a late-time integrated
effect, to the shape of the spectrum tends to overestimate the spectral tilt. For example, analysis of a class of
CDM models (Hancock et al. 1998) (CDM and SCDM,
a subset of the models considered here) finds a spectral
tilt ns = 1.1 0.1. We conservatively restrict the spectral index of the primordial adiabatic density perturbation spectrum, with P (k) k ns , to lie in the interval
ns [0.8, 1.2]. Note that inflation generically predicts
ns 1, with ns slightly less than unity preferred by inflaton potentials which naturally exit inflation.
Small Angle CMB: Dramatic advances in cosmology
are expected in the near future, when the MAP and Planck
satellites return high resolution maps of the CMB temperature and polarization anisotropy. When the measurements are analyzed, we can expect that the best determined cosmological quantities will be the high multipole
C moments, such that any proposed theory must first
5 www.cita.utoronto.ca/knox/radical/bpdata.html
6 www.sns.ias.edu/max/cmb/experiments.html

explain the observed anisotropy spectrum. At present,

however, there is ample CMB data which can be used to
constrain cosmological models.
We take a conservative approach in applying the small
angular scale CMB data as a model constraint. Our intention is to simply determine which quintessence models are
consistent with the ensemble of CMB experiments, rather
than to determine the most likely or best fitting model; examining Figure 5, the error bars are clearly so large that
a best fit has little significance. We have restricted our
attention to that subset of published CMB experimental
data which satisfies the following objective criteria: multifrequency; positive cross-correlation with another experiment; careful treatment of foregrounds. (While we advocate these criteria, our conclusions are not strongly sensitive to this selection of data.) Hence, we use the bandpower estimates from COBE (Bennett et al. 1996), Python
(Platt et al. 1997), MSAM (Cheng et al. 1997), QMAP
(Devlin et al. 1998, Herbig et al. 1998, Oliveira-Costa et
al. 1998), Saskatoon (Netterfield et al. 1997), CAT (Scott
et al. 1996), and RING5M (Leitch et al. 1998) experiments
as the basis of the small angle CMB cosmological constraint. Figure 5 shows the bandpower averages at the effective multipole number, le , with several QCDM models
for comparison. We apply a simple 2 test with the predicted bandpower averages, Tle (defined in Bond 1995).
(A compilation of bandpower averages and window functions is available from Knox5 and Tegmark6 .) Since the
reported bandpower errors are typically not Gaussian distributed, treating the Tle as a Gaussian random variable
can introduce a bias in the estimation of the quality of
agreement. In this case, we also consider a 2 test in the
quantity ln(Tl2e ), following Bond et al. 1998.
Figure 5 illustrates features relevant to limits on the
spectral index, ns . The COBE limit on the spectral index
is reported to be ns = 1.2 0.3 (Gorski et al. 1996, Hinshaw et al. 1996) based on comparison between data and
standard CDM models. In the Figure, though, it is apparent that there is negligible difference in the low angular
scale predictions despite a range of n = 0.2. The difference in the large angular scale integrated Sachs-Wolfe
contribution compensates for the difference in spectral index. Hence, as will be addressed in a later paper, the
COBE limit on ns is somewhat expanded when QCDM is
included.
4. CONCORDANCE RESULTS

We have evaluated the cosmological constraints for the

set of quintessence models occupying the five dimensional
parameter space: w, m , b , h, ns . The results are best
represented by projecting the viable models onto the
m h and m w planes.
The concordance region due to the suite of low red shift
constraints, including the COBE normalization and tilt
ns , are displayed in Figures 6, 7. Each point in the shaded
region represents at least one model in the remaining three
dimensional parameter space which satisfies the observational constraints.
In Figure 6, the boundaries in the m direction are determined by the combined BBN and BF constraints as a

WANG ET AL.

function of h, while h is only restricted by our conservative allowed range and the age constraint. The age does
not impact the m h concordance region, since for the
allowed values of m and h, there is always a model with
a sufficiently negative value of w to satisfy the age constraint. Relaxing either the BBN or BF constraint would
raise the upper limit on the matter density parameter to
allow larger values of m . This requires a simultaneous reduction in the spectral index, ns , in order to satisfy both
the COBE normalization and cluster abundance.
In Figure 7, the upper and lower bounds on m are
again determined by the combination of BBN, BF, and
h. The lower bound on m due to the combination of
the BBN and BF constraints can be relaxed if we allow
a more conservative range for the baryon density, such as
0.006 < b h2 < 0.022 (Levshakov 1998, Olive, Steigman
& Walker 1999). However, the constraints due to 8 and
the shape of the mass power spectrum take up the slack,
and the lower boundary of the concordance region is relatively unaffected. The lower bound on m near w = 1
is determined in part by the shape test; the mass power
spectrum in a model with low m and strongly negative
w is a poor fit to the shape of the APM data, based on a
2 -test. This constraint on models near w = 1 is relaxed
if we allow anti-bias (b < 1), although b < 1 is strongly
disfavored on a theoretical basis. At the other end, for
w >
0.6, the lower bound on m is determined by the
combination of the upper bound on the spectral index, and
the x-ray cluster abundance constraint on 8 . If we further
restrict the bias to b < 1.5, a small group of models at the
upper right corner with w >
0.4 will fail
0.2 and m >
the shape test.
We see that models occupying the fraction of the parameter space in the range 1 w <

0.2 and 0.2 <

m <
0.5 are in concordance with the basic suite of observations,
suggesting a low density universe. It is important to note
that the set of viable models spans a wide range in w; the
concordance region is not clustered around w = 1, or
, but allow such diverse behavior as w 1/3. However, the case w = 0, which can result from the scaling
exponential potential (Ratra & Peebles 1988, Peebles &
Ratra 1988), is clearly in contradiction with observation:
the m required by the x-ray cluster abundance constraint
is incompatible with the matter density parameter allowed
by the BF and BBN constraints. Hence, the models with
w = 0 explored in Ferreira & Joyce 1997, Ferreira & Joyce
1998 are not viable.
We have taken the attitude in our work that current
observational uncertainties are dominated by systematic
errors, so that a conservative method of combining observational constraints is by concordance. We apply the 2
limits for each individual observation to pare down the
viable parameter range. However, it is interesting to compare this to what a naive maximum likelihood estimate
(treating the errors as gaussian) would give. In Figure 8
we show the 2 contour in the m w plane, where the
remaining parameters have been marginalized. This parameter region is only slightly smaller than that resulting from concordance. It is reassuring that this technique
yields approximately the same result, although one should
be cognizant of some of the pitfalls of both methods as
discussed in the Appendix. In using the maximum likelihood technique, we lose some of our ability to identify

Vol. 530

the constraint dominating a particular portion of the contour. However, we show that by lifting the shape test, the
constraint relaxes on the range of m allowed for models with w closer to 1, including CDM. For the sake
of argument, the best fit models, where the likelihood is
maximized, are also shown. We see more clearly that the
shape test drives the preferred models away from w = 1,
towards w 1/2.
The concordance approach offers no such best fit, as
it contains no procedure for weighting or combining data.
However, in Figure 9 we carry out the exercise of artificially shrinking all the error bars, to find the last remaining
models. This is equivalent to imagining that all measurements have accurately determined the intended quantity,
but overstated their uncertainties. This procedure narrows down to the same set of models, m 0.33, but is
not driven as strongly by the shape test as is the maximum
likelihood procedure.
The most potent of the intermediate red shift constraints is due to type 1a supernovae, which we present
in Figure 10. In addition to the SCP results, the HZS
group has presented two different analyses of their catalog
of SNe, based on multi-color light curve shapes (MLCS)
and template fitting; hence we show three SNe results.
Carrying out a maximum likelihood analysis, all three give
approximately the same result for the location of the 2
bound, favoring concordant models with low m , and very
negative w. It is interesting to observe that the SNe bound
is consistent with the core of the low red shift concordance
region, displayed earlier in Figure 9. Based on the SCP
maximum likelihood analysis, Perlmutter et al. 1999 have
reported a limit w 0.6 at the 1 level. A 2 analysis
of the same data gives a somewhat different result: the
Fit C SCP data and the HZS data sets give comparable,
although weaker, results to the likelihood analysis. In the
spirit of conservativism, we have used the weakest bound
which we can reasonably justify. Hence, for the concordance analysis, we use the 2 contour resulting from a 2
test.
The statistical rate of gravitational lensing provides a
counter to the trend towards low matter density. In Figure 11 we present the results of our analysis of the HSTSSS lenses. Other groups have come to similar conclusions,
based on this and other lens surveys. Our results are in
excellent agreement with Torres & Waga 1996, Waga &
Miceli 1999, as well as the more sophisticated analyses carried out by Maoz & Rix 1993, Kochanek 1995, Kochanek
1996, Falco et al. 1998 for CDM. In general, there are
fewer lenses observed than expected based on the volumered shift relation for a low density, -dominated universe.
The disparity between theory and observation is reduced
as the matter density increases, or as w increases.
We have evaluated the x-ray cluster evolution constraint
using the observed abundance of rich clusters at z 1.
This test constrains the amplitude of mass fluctuations
and the rate of perturbation growth. While it has been
argued that this reduces to a bound on 8 for CDM
models (Bahcall & Fan 1998), it has been shown that
the bound depends on w for QCDM scenarios by Wang
& Steinhardt 1998. In Figure 12 we show the consequence of the cluster evolution constraint on the concordance models in the m w plane; at this stage, the early
formation of structure implied by the observations argues

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

against concordant quintessence models with an equationof-state w >

0.3. When the measurements comprising
this constraint improve, we can expect a much more stringent result. Considering the hypothetical situation that
future observations successfully reduce the systematic uncertainty to the present 1 level, the constraint boundary
would shift to the small region with 0.8 <
w<
0.5 and
<
0.3.
0.25 <

We have evaluated the high red shift constraint due to

the select ensemble of CMB anisotropy measurements, using the COBE, Python (Platt et al. 1997), MSAM (Cheng
et al. 1997), QMAP (Devlin et al. 1998, Herbig et al.
1998, Oliveira-Costa et al. 1998), Saskatoon (Netterfield et
al. 1997), CAT (Scott et al. 1996), and RING5M (Leitch et
al. 1998) results. Based on a 2 test in Tl , the set of concordant models projected down to the m h and m w
planes is unchanged from the low red shift concordance region at even the 1 level. This null result from the CMB
should not be too surprising; the current observational
data is capable only of discerning a rise and fall in power in
the C spectrum across 100 300. The results are unchanged if we include additional current CMB results, or
use a 2 test in ln(Tl2 ), as suggested in Bond et al. 1998.
Rather, we must wait for near-future experiments which
have greater coverage, e.g. BOOMERANG, MAT, and
MAXIMA, which are expected to significantly reduce the
uncertainties.
Since the submission of this manuscript, the data
from the MAT (Miller et al, 1999) and BOOMERANG
(Mauskopf et al, 1999) experiments have been released.
However, neither significantly changes our results.
Thus far we have applied the low red shift constraints
in sequence with one of the other intermediate or high
red shift constraints. It is straight forward to see how the
combined set of constraints restrict the quintessence parameter space. Taking the low red shift constraint region,
which is shaped primarily by the BF, BBN, H, and 8 constraints, the dominant bounds on the m w plane are
then due to SNe and lensing. The SNe drives the concordance region towards small m and negative w; the lensing
restricts low values of m . Putting these all together, an
ultimate concordance test is presented in Figure 13. We
see that the resulting concordance region in the m w
plane is very similar to the core region obtained in Figure 9. If the present observations are reliable, we may
conclude that these models are the most viable among the
class of cosmological scenarios considered herein.
It is beyond the scope of the present work to determine how well future observations will determine the values of cosmological parameters in a QCDM scenario. However, we are in a position to highlight those observations
which appear well suited to testing the quintessence hypothesis. Clearly, the first goal must be to distinguish
Q from (Huey et al. 1998). Observations which measure the growth of structure at intermediate red shifts
(z 0.5 1.0) are best suited for the purpose. In this
red shift regime, structure growth is still occurring for the
model, but has shut off significantly for the Q model.
The Ly- determination of the mass power spectrum amplitude is at too high of a red shift to serve this purpose: at
z 2.5, evolution is still matter dominated for m = 0.3
models with w <
1/3. Cluster abundances at z 0.5 1
and the supernovae magnitude-red shift relation are better

suited to this goal.

The shape of the mass power spectrum may prove to
be a strong test of the QCDM scenario, if the observation of a turnover in the power spectrum near k
0.02 0.06 h/Mpc and a break in the slope at higher wave
numbers (Gaztanaga & Baugh 1998) bears out. It may
prove difficult for the simplest, scalar field quintessence,
or for that matter, to generate such a feature. This is
the subject of another investigation (Zlatev et al. 1999).
A test of the tracker quintessence scenario can be made
by determining the change in the equation-of-state. If the
equation-of-state can be measured at the present and at an
earlier epoch, say z 1, we can obtain a crude measure of
the slope, dw/dt. Trackers have the special property that
the equation-of-state becomes more negative at late times:
w 1 as Q 1. A measurement of dw/dt > 0 would
argue against tracker quintessence.
More exotic observational tests can be used to discover
the presence of a quintessence field. For example, if the
quintessence field is coupled to the pseudoscalar F F of
electromagnetism as suggested by some effective field theory considerations (Carroll 1998), the polarization vector
of a propagating photon will rotate by an angle that
is proportional to the change of the field value Q along
the path. CMB polarization maps can potentially measure
the from red shift 1100 to now (Lue et al. 1999) and
distant radio galaxies and quasars can provide information of from red shift a few to now (Carroll 1998). If
these two observations generate non-zero results, they can
provide unique tests for quintessence and the tracker hypothesis, because tracker fields start rolling early (say, before matter-radiation equality) whereas most non-tracking
quintessence fields start rolling just recently (at red shift
of a few).
5. CONCLUSIONS

We have applied a battery of tests and constraints to

the family of quintessence cosmological models, determining the range of parameters which are concordant with
observations. The most reliable constraints are those resulting from low red shift observations, and the COBE
normalization of the mass power spectrum. These restrict
QCDM models to a narrow range of parameters, characterized by low matter density, 0.2 m 0.5, and
negative equation-of-state, 1 w <
0.2. While the
intermediate red shift results are still developing, the implications are very exciting. The SNe observations narrow
the range of matter density near m 0.3 0.4, and force
the equation-of-state to w <
0.4. While this appears
consistent with the core of the low red shift concordance,
the potential for conflict is present if the matter power
spectrum shape test demands w >
1. Our results based
on low red shift observations are given by Figures 6, 7;
adding the supernovae constraints, which are more recent
and whose systematic errors have not been fully tested,
produces the narrower range shown in Figure 13.
To what degree do current uncertainties in the Hubble parameter, the spectral tilt and other cosmic parameters obstruct the resolution in w? To judge this issue,
we have performed an exercise in which we fix h = 0.65,
b h2 = 0.019, and we choose the spectral tilt to insure
that the central values of the COBE normalization and

10

WANG ET AL.

the cluster abundance constraint are precisely satisfied.

In Figures 14 and 15, we show how different constraints
restrict the parameter planes. Note first the long, white
concordance region that remains in the m w plane,
which is only modestly shrunken compared to the concordance region obtained when current observational errors
are included. The region encompasses both and a substantial range of quintessence. Hence, current uncertainties in other parameters are not critical to the uncertainty
in w. The figure further shows how each individual constraint acts to rule out regions of the plane. The color or
numbers in each patch represent the number of constraints
violated by models in that patch. It is clear that regions
far from the concordance region are ruled out by many
constraints. Both figures also show that the boundaries
due to the constraints tend to run parallel to the boundary of the concordance region. Hence, shifts in the values
or the uncertainties in these measurements are unlikely to
resolve the uncertainty in w by ruling out one side or the
other either the constraints will remain as they are, in
which case the entire concordance region is allowed, or the
constraints will shift to rule out the entire region.
New measurements not represented in this figure will
be needed to distinguish from quintessence. At this
point, high precision measurements of the cosmic microwave background anisotropy are the most promising.
The results presented in this paper apply to quintessence
models in which the equation-of-state is constant or slowly
varying with time. In the latter case, setting w = w,
e given
in equation (3), gives an excellent approximation to the
observational predictions for a broad class of models.
Particularly important classes of quintessence models
are tracker and creeper fields. The tracker models are
highly appealing theoretically because they avoid the
ultra-fine tuning of initial conditions required by models with a cosmological constant or other (non-tracking)
quintessence models. An additional important feature of
these models is that they predict a definite relationship be-

Vol. 530

tween the present day energy density and pressure, which

yields a lower bound on the constant, effective equationof-state, near w
e 0.8 (Steinhardt et al. 1999). Note
that the effective or averaged equation-of-state as computed from Eq. (3) is about 10 per cent larger than the
value of w today (given in Table I). In Figure 16 we add
this bound to the low red shift constraints, obtaining the
concordance region for tracker quintessence. This region
retains the core of our earlier low red shift concordance,
and is consistent with the SNe constraints. Creeper fields
occur for the same potentials as trackers, but the initial energy density exceeds the radiation density at early times.
The consequence is that the field rapidly rolls down the
potential, towards a point which is only mildly (logarithmically) sensitive to the initial conditions, where it sticks
and is effectively frozen with constant potential energy at
very early times. Hence, the creeper field has an equationof-state w = 1, and is effectively indistinguishable from
a cosmological constant today.
In Figure 17 we combine all current observations on
tracker models.
Since these are arguably the bestmotivated theoretically, we identify from this restricted
region a sampling of representative models, listed in Table I, with the most attractive region for quintessence models being m 0.33 0.05, effective equation-of-state
w 0.65 0.07 and h = 0.65 0.10 and are consistent with spectral index ns = 1 indicated by the dark
shaded region in Figure 17. These models represent the
best targets for future analysis. The challenge is to prove
or disprove the efficacy of these models and, if proven, to
discriminate among them.
We wish to thank Neta Bahcall, John Peacock and
Michael Strauss for useful discussions. This research was
supported by the US Department of Energy grants DEFG02-92ER40699 (Columbia) and DE-FG02-91ER40671
(Princeton).

APPENDIX

Throughout this paper, we have chosen to judge models by combining observational constraints according to the
concordance method in addition to the maximum likelihood estimator (MLE) method. We have asserted that the
concordance method is conservative, sometimes giving a more reliable judge of the situation than the MLE method,
especially when the observational constraints may be dominated by systematic, nongaussian, and/or correlated errors.
We advocate using both concordance and MLE methods, as we have done in this paper, and then analyzing the source of
any discrepancy before determining which models should be ruled out. Since it has been commonplace to present MLE
results alone, we thought it would be useful to illustrate some of the pitfalls that can arise.
For this purpose, we employ a toy example in which we have two parameters, A and B, and two independent (observational) constraints, represented as the 2 regions C1 and C2 , which restrict the allowed ranges of parameters. This is
meant to be a simplification of our real situation where we have a five-dimensional parameter space to analyze quintessence
models, and we have many observational constraints. In our paper, we have tried to determine constraints in the m -w
plane by projecting, effectively, from five-dimensions to two. In our toy model, we imagine projecting onto the A-axis
to determine the constraint on A, indicated as a bar along the axis. Our interest is to compare the concordance region
corresponding to 2 with the 95%CL contour from the MLE method.
We present several simple examples in which there is a large disparity between the concordance and MLE procedures.
In Figures 19, 18, we represent a two dimensional parameter space, A-B, with two independent constraint regions, C1,2
shown as shaded rectangles. The projection of the concordance and MLE regions, Cconc and CMLE respectively, onto
the horizontal axis are shown as thick strips. Although the following discussion is qualitative, the relative sizes of the
projected strips are correct.
Given the two observations, C1 and C2 , the concordance region for A is obtained by: (a) finding their intersection in
two-dimensions; and (b) projecting the two-dimensional intersection Cconc onto the A-axis to obtain a bar. Note that we
do not project first, and then take the intersection. This method can lead to gross errors. For example, consider Figure

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

11

19, in which the C1 and C2 have no intersection at all. By our method, the concordance region is properly identified as
the null set, whereas projecting first and then finding the intersection would produce a considerable band of acceptance,
a false conclusion.
According to the MLE method, we are required to know the central value,
~ , of each region, and assume the errors, ~ ,
are Gaussian. In the following, we make the simplistic assumptions that the likelihood function for each observation is
symmetric about the center of the constraint region, C1 or C2 . We weight each point ~x by a Gaussian f (~x, ~, ~) for each
of
and identify CMLE , the contour of constant f1 f2 or 2 which contains 95% of the total probability,
R
R the two constraints,
f f d~x = 0.95 f1 f2 d~x. This is a straightforward procedure.
C 1 2
Case 1: Figure 18 illustrates a case where C1 encompasses C2 . In this situation, the MLE method (indicated by
the lower bar) produces a smaller acceptance region than the concordance method. If the errors are truly gaussian
and uncorrelated, the conclusion based on MLE is the better representation of uncertainty. Note that the concordance
region includes the MLE region plus an additional range of A; so, the error made with concordance is to include too
much. However, no model is ruled out by concordance which intuition suggests ought not be eliminated. We call this
conservative, and our aim is to find a robust, conservative method.
The MLE is not a robust, conservative method, as can be illustrated by the same figure. Suppose that the observations
in C1 and C2 are suspected to be nongaussian or systematic or correlated. Then, we are clearly mistaken to rely on the
MLE method and eliminate the range of A which lies within the concordance region but outside the MLE region. In
this example, the difference is only modest, but if we were combining a number of observations, the MLE allowed region
would be much tinier than the concordance region, and the error in trusting the MLE method would be more serious.
This point is directly relevant to this paper.
Case 2: The observational constraints C1 and C2 intersect in a small region or, as illustrated in Figure 19, have no
intersection at all. This case is opposite to Case 1 in that the MLE method produces a larger acceptance region than the
concordance method. For example, the concordance region in Figure 19 is the null set, whereas the MLE contour suggests
a large acceptance region. This is not a case where the MLE method is being conservative; rather, it is a case where
the MLE method is misleading. Intuition dictates that the observational constraints are in conflict, and the concordance
method reflects this conclusion by producing a null concordance region. The MLE method, taken at face value, suggests
a broad range of agreement. To be fair to the MLE methodology, one is not supposed to accept the 95% MLE contour
at face value. The contour represents a probability compared to the maximum likelihood point, and one is supposed to
check that this point is indeed a good fit. In practice, though, this step is often ignored or discounted. For example, if
the maximum likelihood point has a high value of 2 by the conventional 2 -test, this is often (properly) considered a
problem due to underestimating experimental errors. But, as indicated by this example, the same statistical result can
be an indication that there is a true contradiction between models and data, and that completely new models need to
considered. Hence, a contradiction between concordance and MLE methods is a warning to examine closely the cause.
Case 3: Suppose constraint C2 is obtained by combining many measurements with small statistical uncertainty but
unknown correlated error. Then, in a MLE analysis, constraint C2 receives undue statistical weight.
If we drop the simplistic assumption that the likelihood function for each individual constraint is symmetric about the
center of the constraint region, which is rare, other kinds of discrepancies between concordance and maximum likelihood
can occur. For example, the MLE region may be shifted with respect to the concordance region so that each test allows
models which the other does not.
As a prominent example, the current, highly provocative measurements of type 1a supernovae by the High-Z Supernova
Search Team (HZS: Riess et al. 1998, Garnavich et al. 1998) and the Supernova Cosmology Project (SCP: Perlmutter
et al. 1998) exemplify all three cases above. For the HZS data based on MLCS (multi-color light curve shape) analysis
or for SCP using the Fit C selected data set Perlmutter et al. 1998, the concordance region is significantly larger than
the MLE acceptance region, as demonstrated by comparing Figures 10 and 13. So, relying on the MLE 95%CL region
eliminates models which formally pass the absolute 2 test at the 95% level. While statisticians may argue that the MLE
likelihood estimate is more reliable assuming uniform prior over the parameter space, in comparing qualitatively different
models (which do not have uniform prior), the readers should beware that MLE can potentially rule out an entire model
even if the model agrees at better than the 95% confidence level (as judged by 2 ). Especially at this early stage when
better data will soon be available, we advocate the more cautious, concordance approach. This first example is like Case
1. We have also noted that the scatter in the supernova red shift - magnitude data is so wide that no model which we
have tested passes a 2 test with the SCP full data set. Hence, this is an example, like Case 2, where the concordance
region is null but the MLE acceptance region is large. Finally, in a MLE or Fisher matrix analysis which combines the
supernovae measurements with other observations, the fact that there are many individual supernovae with small reported
uncertainty gives these measurements heavy statistical weight. However, as the survey teams admit, the measurement
approach is new and there remains the possibility that as yet unidentified physical effects cause a systematic, apparent
reddening of the data. As in Case 3, results obtained by simple statistical combination of supernovae data with other
measurements should be viewed cautiously.
Another feature of maximum likelihood analysis, well-known to practitioners but perhaps unappreciated by some, is
the fact that the estimation of parameters in a multi-dimensional fit can be markedly different from the estimate when
marginalizing over some parameters. For example, the range of w spanned by the MLE 95% confidence region in the
m -w plane (as considered by Perlmutter et al. 1999 and in this paper) is significantly narrower than the MLE 95%
confidence region in the three-dimensional parameter space including h and significantly broader than the MLE 95%
confidence region obtained by marginalizing over m and collapsing to a one-parameter fit to w. (This is due to the fact

12

WANG ET AL.

Vol. 530

that 2 criterion for 95% confidence depends on the dimensionality of the parameter space.)
These simple cases demonstrate the differences in the concordance and MLE procedures and, especially, some problems
which can arise in MLE analysis. Because we maintain the position that systematic uncertainties dominate the errors in
constructing the constraint regions, we advocate the concordance approach as being more conservative for cosmological
analysis at the present time. In general, caution must be exercised when the two methods disagree significantly, and the
source of the discrepancy must be understood in order to determine the true acceptance region.
REFERENCES
Affleck, I. et al. 1985, Nucl Phys B, 256, 557
Bahcall, N. A. & Fan, X. 1998, ApJ, 504, 1
Bahcall, N. A., Fan, X., & Cen, R. 1997, ApJ, 485, L53
Banks, T. 1996, hep-th/9601151
Bardeen, J. 1986, in InnerSpace/OuterSpace, ed. Edward W. Kolb,
Michael S. Turner, David Lindley, Keith Olive, & David Seckel
(Chicago: University of Chicago Press), 212
Barreiro, T., Carlos, B. & Copeland, E. J. 1998, Phys. Rev. D, 57,
7354
Bartelmann, M. et al. 1998, Astron. Astrophys., 330, 1
Bennett, C. L., et al. 1996, ApJ, 464, L1
Bertschinger, E. & Dekel 1989, ApJ, 335, L5
Binetruy, P., Gaillard, M. K. & Wu, Y.-Y. 1996, Nucl Phys B, 481,
109
Binetruy, P. 1999, Phys. Rev. D, 60, 063502
Blanton, M., Cen, R. Ostriker, J. P., & Strauss, M. 1998, ApJ, 522,
590
Bond, J. R. 1995, Astrophys Lett & Comm, 32, 63
Bond, J. R., Jaffe, A. H., & Knox, L. 1998, astro-ph/9808264
Bunn, E. F. & White, M. 1997, ApJ, 480, 6
Burles, S. & Tytler, D. 1997, ApJ, 499,699
Burles, S. & Tytler, D. 1997, ApJ, 507, 732
Burles, S. & Tytler, D. 1998, SSRv, 84, 65B (astro-ph/9712265)
Caldwell, R. R., Dave, R., & Steinhardt, P. J. 1998, Phys. Rev. Lett.,
80, 1582
Caldwell, R. R., & Steinhardt, P. J. 1998, Phys. Rev. D, 57, 6057
Carlberg, R. G., Morris, S. M., Yee, H. K. C., & Ellingson, E. 1997,
ApJ, 479, L19
Carroll, S. 1998, Phys. Rev. Lett., 81, 3067.
Cen, R. & Ostriker, J. P. 1998, astro-ph/9809370
Chaboyer, B. et al. 1998, ApJ, 494, 96
Cheng, Y.-C. & Krauss, Lawrence M. 1999, ApJ, 514, 25.
Chiu, W., Ostriker, J.P. and Strauss, M.A. 1998, ApJ, 494, 479
Cheng, E. S. et al. 1997, ApJ, 488, L59
Coble, K. et al. 1997, Phys. Rev. D, 55, 1851
Croft, R. A. C. et al. 1998, ApJ, 495, 44
Croft, R. A. C. et al. 1998, ApJ, 520, 1
da Costa, L.N. et al. 1998, MNRAS, 299, 425
Dave, R. 1998, PhD. Thesis, University of Pennsylvania
Davis, M. et al. 1985, ApJ, 292, 371
Davis, M., Nusser, A., & Willick, J. 1993, ApJ, 473, 22
Dekel, A. et al 1999, ApJ, 522, 1
Devlin, M. J. et al. 1998, ApJ, 509, L69
Evrard, August E. 1997, MNRAS, 292, 289
Falco, E. E. et al. 1997, ApJ, 484, 70
Falco, E. E., Kochanek, C. S., & Munoz, J. A. 1998, ApJ, 494, 47
Ferreira, P. G. & Joyce, M. 1997, Phys. Rev. Lett., 79, 4740
Ferreira, P. G. & Joyce, M. 1998, Phys. Rev. D, 58, 023503
Freedman, W. L. et al. 1998, in IAU Symp. 183, Cosmological
Parameters and the Evolution of the Universe, ed. K. Sato (astroph/9801080)
Freudling, W. et al. 1999, ApJ, 523, 1
Frieman, J. et al., 1995, Phys. Rev. Lett., 75, 2077
Fry, J. N. 1986, ApJ, 461, L65
Fukugita, M., Hogan, C. J., & Peebles, P. J. E. 1997, ApJ, 503, 518
Gardner, J. P., et al. 1997, ApJ, 486, 42
Garnavich, P. M. et al. 1998, ApJ, 509, 74
Gaztanaga, E., & Frieman, J. A. 1994, ApJ, 437, L13
Gaztanaga, E. & Baugh, C. M. 1998, MNRAS, 294, 229
Giovanelli et al, 1998, ApJ, 505, L91
Gorski, K. M. et al. 1996, ApJ, 464, L11
Gorski, K. M. et al. 1998, ApJ, 114, 1
Gott, J. R., III 1982, Nature 295, 304
Hamuy, M. et al. 1996, AJ, 112, 2398
Hancock, S. et al. 1998, MNRAS, 294, L1
Herbig, T. et al. 1998, ApJ, 509, L73
Hill, C. & Ross, G. G. 1988, Nuc Phys B, 311, 253
Hill, C. & Ross, G. G. 1988, Phys Lett B, 203, 125
Hinshaw, G. et al. 1996, ApJ, 464, L25
Hu, W. 1999, ApJ, 506, 485
Hudson, M.J. et al. 1999, ApJ, 512, L79
Huey, G., Wang, L., Dave, R., Caldwell, R. R., & Steinhardt, P. J.
1999, Phys. Rev. D, 59, 063005
Huey, G. & Steinhardt, P. J. 1999, in preparation.

Hui, L. 1999, ApJ, 516, 519

Hui, L., Stebbins, A. & Burles, S. 1999, ApJ, 511, 5
Jenkins, E., et al. 1999, ApJ, 520, 182
Kaiser, N. 1986, in InnerSpace/OuterSpace, ed. Edward W. Kolb,
Michael S. Turner, David Lindley, Keith Olive, & David Seckel
(Chicago: University of Chicago Press), 258
Keeton, C. R. & Kochanek, C. S. 1997, ApJ, 487, 42
Kochanek, C. S. 1995, ApJ, 453, 545
Kochanek, C. S. 1996, ApJ, 466, 638
Krauss, L.M. & Turner, M.S., Gen Rel Grav 27, 1137
Kundic, T. et al. 1997, AJ, 114, 507
Lauer, T. & Postman, M. 1994, ApJ, 425, 418
Lauer, T. et al. 1998, ApJ, 499, 577
Leitch, E. M. et al. 1998, astro-ph/9807312
Levshakov, S. 1998, astro-ph/9808295
Linde, A. 1999, Phys. Rev. D, 59, 023503
Lubin, L. et al. 1996, ApJ, 460, 10
Lue, A., Wang, L. & Kamionkowski, M. 1999, Phys. Rev. Lett., 83,
1506
Malhotra, S., Rhoads, J. & Turner, E. 1997, MNRAS, 288, 138
Maoz, D. et al 1993, ApJ, 409, 28
Maoz, D. & Rix, H.-W. 1993, ApJ, 416, 425
Mauskopf, P.D., et al. 1999, astro-ph/9911444.
McDonald, P. & Miralda-Escude, J. 1998, ApJ, 518, 24
McGaugh, S. 1998, ApJ, 499, 41
Miller, A.D., et al. 1999, ApJ, 524, L1.
Mo, H. J. & Fukugita, M. 1996, ApJ, 467, L9
Myers, S. T. et al. 1997, ApJ, 492, 110
Netterfield, B. et al. 1997, ApJ, 474, 47
Olive, K., Steigman, G. & Walker, T. 1999, astro-ph/9905320
Oliveira-Costa, A. et al. 1998, ApJ, 509, L77
Ostriker, J. P. & Steinhardt, P. J. 1995, Nature 377, 600
Paczynski, B., private communication
Peacock, J. 1997a, MNRAS, 284, 885
Peacock, J. 1997b, private communication
Peebles, P. J. E. & Ratra, B. 1988, ApJ, 325, L17
Perlmutter, S. et al 1998, ApJ, 517, 565
Perlmutter, S., Turner, M. S., and White, M., Phys. Rev. Lett., 83,
670
Platt, S. R. et al. 1997, ApJ, 475, L1
Press, W. H. & Schechter, P. 1974, ApJ187, 425
Ratra, B. & Peebles, P. J. E. 1988, Phys. Rev. D, 37, 3406
Ratra, B. & Quillen, A. 1992, MNRAS, 259, 738
Riess, Adam G. et al. 1998, AJ, 116, 109
Salaris, M. & Weiss, A. 1998, Astron. Astrophys., 335, 943
Schechter, P. L. et al. 1997, ApJ, 475, L85
Schramm, D. N. & Turner, M. S. 1998, Rev Mod Phys, 70, 303
Scott, P. F. et al. 1996, ApJ, 461, L1
Sigad, Y. et al. 1998, ApJ, 495, 516
Silveira, V. & Waga, I. 1997, Phys. Rev. D, 56, 4625
Seljak, U. & Zaldarriaga 1996, ApJ, 469, 437
Spergel, D. & Pen, U.-L. 1996, ApJ, 491, L67
Steidel, C. C. et al. 1998, ApJ, 492, 428
Steinhardt, P. J., Wang, L., & Zlatev, I. 1999, Phys. Rev. D, 59,
123504
Strauss, Michael 1997, in Critical Dialogues in Cosmology, ed. Neil
Turok (Singapore: World Scientific), 423
Tammann, G. A. 1998, in Proceedings of the 8th Marcel Grossman
Syposium, ed. T. Piran (Singapore: World Scientific) (astroph/9805013)
Tegmark, M & Peebles, P. J. E. 1998, ApJ, 500, L79
Bloomfield-Torres, L. F. & Waga, I. 1996, MNRAS, 279, 712
Turner, E. L., Ostriker, J. P. & Gott, J. R. III 1984, ApJ, 284, 1
Turner, M. S. & White, M. 1997, Phys. Rev. D, 56, R4439
Waga, I. & Miceli, A. P. M. R. 1999, Phys. Rev. D, 59, 1035
Wambsganss, J., Cen, R., & Ostriker, J. P. 1998, ApJ, 494, 29
Wang, L. & Steinhardt, P. J. 1998, ApJ, 508, 483
Weinberg, D. H. et al. 1998, ApJ, 522, 563
Weiss, N. 1987, Phys. Rev. Lett., 197, 42
Wetterich, C. 1995, Astron. Astrophys., 301, 321
White, S. D. M. et al. 1993, Nature, 366, 429
Willick, J. et al. 1997, ApJ, 486, 629
Willick, J. & Strauss, M. 1998, ApJ, 507, 64
Wu, X.-P. & Mao, S. 1996, ApJ, 463, 404

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

Zaroubi, S. et al. 1997, ApJ, 486, 21

Zehavi, I. 1998, PhD Thesis, Hebrew University
Zlatev, I., Wang, L., & Steinhardt, P. J. 1998, Phys. Rev. Lett., 82,
896

13

Zlatev, I., Wang, L., Caldwell, R. R., & Steinhardt, P. J. 1999, in

preparation

14

WANG ET AL.

Vol. 530

120

60

P(k)/ (h Mpc )
8

80

-3

100

40

20

0
0.00

0.05

-1

0.10

0.15

k (h Mpc )
Fig. 1. The solid circles with 1 error bars are the Peacock APM data we use to test the power spectra shape. The five solid lines are
quintessence models that pass the shape test with a confidence level of 95%. The model parameters for the black, blue, red, green, and purple
curves are: w = -1, -1/2, -1/3, -1/6, 0; Q = 0.70, 0.60, 0.55, 0.43, 0.20; = 0.20, 0.26, 0.29, 0.37, 0.52. The bias factor is optimized for each
model shown in order to obtain the minimum 2 : b = 1.01, 1.24, 1.46, 1.81, 2.49, respectively. All models have ns = 1 and b h2 = 0.02. For
comparison, the dashed line is a standard CDM model with best-fit bias 0.8 which fails the shape test.

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

15

bulk velocity [km/s]

1000

model 1
model 2
model 3
model 4
model 5

800

600

400

200

50

100

150

200

top hat radius [Mpc/h]

Fig. 2. The bulk velocity predictions as a function of radius (assuming a top hat window function) are shown for a representative set of
QCDM models, and CDM given at the end of this paper in Table I. Surrounding the best-fit quintessence model (Model 2), we have shown
the shaded sheath corresponding to the 95%CL, according to a Maxwellian distribution of bulk velocities. The diamond, circle, square, and
triangle show the bulk velocities measured by Dekel et al. 1999, Giovanelli et al. 1998, Hudson et al. 1999, Lauer & Postman 1994 with 2
error bars, respectively. We have idealized the window function for the galaxy velocity catalog, assuming a spherical top hat.

16

WANG ET AL.

Vol. 530

2
1

HZS

(mM)

0
1

0.1

0.1

2
1

SCP

0
1

red shift z
Fig. 3. The magnitude - red shift relationship determined by type 1a SNe is shown, for the HZS (using the MLCS analysis method)
above, and SCP (full data set) below. The horizontal (m M ) = 0 reference line shows the prediction of an empty universe (total = 0),
which has been subtracted from all data and theoretical curves. The thin dashed and dot-dashed curves show the predicted magnitude-red
shift relationship for flat models with = 1 and m = 1, respectively. The vertical offset of the data has been determined by minimizing
the 2 to the best fit CDM model with m = 0.3, which is given by the thick, solid curve. The predictions for quintessence (QCDM) models
with w = 2/3, 1/3 for the same matter density are shown by the thick dashed and dot-dashed curves.

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

P(k,z=0) [Mpc/h]

10

17

Non-linear
contributions
distort
power spectrum
at z=0

10

model 1
model 2
model 3
model 4
model 5

10

10

-1

10

-2

10

-3

10

-2

10

-1

10

10

k [h/Mpc]
10

Non-linear
contributions
negligible
at z=2.5

P(k,z=2.5)x(2)3 [km/s]

10

10

10

10

10

10

-5

10

-4

10

-3

k [km/s]

-2

-1

10

10

-1

Fig. 4. The upper panel compares the linear mass power spectrum at z = 0 for the representative CDM and QCDM models in Table I.
All models are COBE normalized and satisfy the cluster abundance constraint on 8 . The solid and dashed curves have ns = 1; the dotted
and dot-dashed curves have ns = 1.2. The shaded region in the top panel indicates where non-linear contributions are non-negligible. The
lower panel shows the same power spectra projected back in time to red shift z = 2.5 and rescaled by the appropriate value of h at red shift
z. Note that, in converting to z = 2.5, the abscissa in the lower panel has been rescaled so that it is expressed in terms of velocity; once this
model-dependent rescaling is made, the models can be compared directly to the data. We show the constraints on the power spectrum, with
1 error bars, as deduced from the Ly- forest. Among our representative models, the ns = 1 models are preferred over the ns = 1.2 models.

18

WANG ET AL.

[l(l+1)Cl/2]

1/2

80

60

Vol. 530

QMAP
PY
MSAM
SK
CAT
RING

40

20

10

100

1000

multipole l
Fig. 5. The bandpower averages used as the basis of the small angle CMB constraint are shown. For comparison, the black, blue, red,
green, and purple curves are models 1-5, given at the end of this paper in Table I. The differences are small, but distinguishable in near-future
experiments. All have acoustic peaks lying significantly above those of the standard cold dark matter model (thin black). The higher results
favor our representative models with ns = 1 over those with ns = 1.2.

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

19

LOW RED SHIFT

0.8

0.6

m
0.4

BBN +

BF

H
H

0.2

0.0
0.4

BBN + B

0.5

0.6

0.7

0.8

0.9

h
Fig. 6. The projection of the concordance region on the m h plane, on the basis of the low red shift observational constraints only,
is shown. The observations which dominate the location of the boundary are labeled.

20

WANG ET AL.

Vol. 530

LOW RED SHIFT + COBE

0.8

0.6
H + BBN + BF

m
0.4

0.2

8+

SHAPE

0.0
-1.0

ns

H + BBN + BF

-0.8

-0.6

-0.4

-0.2

0.0

equation-of-state w
Fig. 7. The projection of the concordance region on the m w plane, on the basis of the low red shift and COBE observational
constraints only, is shown. The observations which dominate the location of the boundary are labelled. If a wider range for the baryon
density is allowed, such as 0.006 < b h2 < 0.022, the shape test and 8 constraint determine the location of the low m boundary, and the
concordance region extends slightly as shown by the light dashed line.

No. 1, 2000

0.8

COSMIC CONCORDANCE AND QUINTESSENCE

21

LOW RED SHIFT + COBE

Maximum Likelihood

0.6

m
0.4

w/shape test
w/o shape test

0.2

0.0
-1.0

-0.8

-0.6

-0.4

-0.2

0.0

equation-of-state w
Fig. 8. The 2 maximum likelihood contours in the m w plane with the low red shift and COBE observational constraints only are
shown. The dotted and dashed curves show the likelihood contours with and without the mass power spectrum shape test. The set of models
which maximize the likelihood in each case are shown by the solid circle and the thick line. The shape test pushes models away from w = 1,
towards w 1/2. The solid line shows the 2 allowed region according to the concordance region for comparison.

22

WANG ET AL.

Vol. 530

LOW RED SHIFT + COBE

0.8

Concordance region
when error range is reduced to:

2
1
0.5
0.25

0.6

m
0.4

0.2

0.0
-1.0

-0.8

-0.6

-0.4

-0.2

0.0

equation-of-state w
Fig. 9. We carry out the exercise of shrinking the error bars on all measurements to obtain the equivalent best fit models in the
concordance approach. The surviving models have m 0.33 and 0.9 < w < 0.7. This is similar to the best fit models obtained from the
maximum likelihood method, however the concordance models are not as strongly affected by the shape test.

No. 1, 2000

0.8

COSMIC CONCORDANCE AND QUINTESSENCE

23

Supernovae Constraint
based on
Maximum Likelihood Estimate

0.6

m
0.4

S Ne

0.0
-1.0

its

0.2 S N
e lo
we
r

upp
er
lim

li m

-0.8

it s

-0.6

-0.4

-0.2

-0.0

equation-of-state w
Fig. 10. The 2 maximum likelihood constraints on the m w plane, due to the SCP (solid), HZS MLCS (short dashed), and HZS
template fitting methods (dot-dashed). The light, dashed line shows the low red shift concordance region.

24

WANG ET AL.

0.8

Vol. 530

Low Red Shift + COBE

+
Gravitational Lens Count Constraint

0.6

m
0.4

0.2

0.0
-1.0

rule
d
len out b
sin
g y

-0.8

-0.6

-0.4

-0.2

equation-of-state w
Fig. 11. The 2 gravitational lensing constraint on the low red shift concordance region is shown.

0.0

No. 1, 2000

0.8

COSMIC CONCORDANCE AND QUINTESSENCE

25

LOW RED SHIFT + COBE

+ Cluster Evolution

0.6

m
0.4

Ev
r
e
t
s
C lu

0.2

0.0
-1.0

-0.8

-0.6

io
l ut

-0.4

-0.2

0.0

equation-of-state w
Fig. 12. The effect of the x-ray cluster abundance evolution constraint on the projection of the concordance region to the m w plane
is shown by the dark shaded region. The light shaded region is due to the low red shift constraints only.

26

WANG ET AL.

Vol. 530

LOW RED SHIFT + CMB

+ SNe + LENSING

0.8

0.6
H + BBN + BF

SN
e

0.4

0.2

lensing

0.0
-1.0

8
H+
BBN + BF

-0.8

-0.6

ns

-0.4

-0.2

0.0

equation-of-state w
Fig. 13. The dark shaded region is the projection of the concordance region on the m w plane with the low, intermediate, and high
red shift observational constraints. The dashed curve shows the 2 boundary as evaluated using maximum likelihood, which is the same as
Figure 10 (See the Appendix for a comparison of the tests and a discussion of the pitfalls of the maximum likelihood approach.)

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

27

Observational constraints assuming

h=0.65, b h2 =0.019, COBE + Cluster abundance constraint
1.0
6
5

0.8

0.6
3

upp SNe
er b
ou

t
spec

nd

ral ti

lt

er
u pp

4
6

baryon fraction upper limit

0.4

cluster abundance evolution uppe

r limit

0.2

fails
exclu
s ha
ded
pe t
by le
est
nsin
2
g

SN
low e
bo er
un
d

-1.0

bulk f

lo

70
<3
wv

km

/s

5
3
4

lower limit based on M/L & LSS

-0.8

0.0

i
lim

sp
lo ect
we ra
r l l til
im t
it

-0.6

-0.4

-0.2

0.0

equation-of-state w
Fig. 14. The concordance region (white) resulting if we artificially set h = 0.65 and b h2 = 0.019 precisely and fix the spectral tilt
to precisely match the central values of COBE normalization and cluster abundance measurements. The curves represent the constraints
imposed by individual measurements. The curves divide the plane into patches which have been numbered (and colored) according to the
number of constraints violated by models in that patch.

28

WANG ET AL.

e
ag

1.0

Vol. 530

>
10
6

r
Gy

shape test u.b.

0.8

spec

tral t

ilt u.

b.

SNe u.b.

0.6

0.4

1
3

0.2

baryon fractio
cluster evolution u.b.

n u.b.

shape test l.b.

lensing l.b.

baryon fraction l.b.

3
4

SNe l.b.

0.0
0.5

0.6

0.7

0.8

h
Fig. 15. The concordance region (white) resulting if we artificially set b h2 = 0.019 and fix the spectral tilt to precisely match the
central values of COBE normalization and cluster abundance measurements. The curves represent the constraints imposed by individual
measurements. The curves divide the plane into patches which have been numbered (and colored) according to the number of constraints
violated by models in that patch.

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

29

LOW RED SHIFT + COBE

for Quintessence

0.8

creeper /
0.6
H + BBN + BF

0.2

H + BBN + BF
0.0
-1.0

-0.8

+n

0.4

t r a ck e r

-0.6

-0.4

-0.2

0.0

effective equation-of-state w
Fig. 16. The concordance region based on COBE and low red shift tests for tracker quintessence is shown. The thin black swath along
w = 1 shows the allowed region for creeper quintessence and . The equation-of-state is time-varying; the abscissa is the effective (average)
w.

30

WANG ET AL.

Vol. 530

LOW RED SHIFT + CMB

+ SNe + LENSING
for Quintessence

0.8
creeper /
0.6

0.2

4
2

5
3

H + BBN + BF
0.0
-1.0

-0.8

SN
e

-0.6

+n

0.4

t r a ck e r l
imit

H + BBN + BF

-0.4

-0.2

0.0

effective equation-of-state w
Fig. 17. The overall concordance region based low, intermediate, and high red shift tests for tracker quintessence is shown. The thin
black swath along w = 1 shows the allowed region for creeper quintessence and . The equation-of-state is time-varying; the abscissa is the
effective (average) w as defined in Eq. (3). The dark shaded region corresponds to the most preferred region (the 2 maximum likelihood
region consistent with the tracker constraint), m 0.33 0.05, effective equation-of-state w 0.65 0.10 and h = 0.65 0.10 and are
consistent with spectral index n = 1. The numbers refer to the representative models that appear in Table I and that are referenced frequently
in the text. Model 1 is the best fit CDM model and Model 2 is the best fit QCDM model.

No. 1, 2000

COSMIC CONCORDANCE AND QUINTESSENCE

31

B
C

concordance
NULL
MLE

Fig. 18. The constraint regions C1 and C2 do not intersect. In the concordance method for determining bounds on A, we first find
the intersection of the C1 and C2 in the full, higher-dimensional parameter space, and then we project that intersection region to obtain the
constraint on A. In this case, the concordance region is null. The MLE method always allows some finite region of 95%CL.

32

WANG ET AL.

Vol. 530

B
C

concordance
MLE

Fig. 19. The constraint regions C1 and C2 overlap. The projection of the concordance and MLE regions onto parameter A, along the
horizontal axis are shown by shaded strips.

No. 1, 2000

tra ker QCDM

-1.0

-0.72

Model Parameters

0.33 0.65 0.041

m

0.35

ns

0.65

1.0

0.039

1.0

Ba kground Evolution Quantities

q0
H0 d j =1
dw=dz
g (
)
0.94 -0.51
1.52
0.0
0.80

t0

H0 t0

13.2

0.88

14.1

L z

-0.20

1.44

0.063

0.74

(
)
0.54
m

0.55

Flu tuation Spe trum Quantities

8
50
v50
v150
+140
4.5
0.90 0.16 290+270
200 150 100
 105

4.1

0.83

0.15

260

+240

180

130

+120
90

+100
-0.60 0.20 0.80 0.025 1.2 11.7 0.96 -0.22
1.48
0.037 0.55
0.40
4.3
0.92 0.17 220+200
150 110+120
80
00
-0.60 0.44 0.60 0.056 1.0 13.4 0.82 0.0
1.36
0.057 0.76
0.63
3.4
0.80 0.13 260+240
130 90
180
+180
+90
-0.41 0.30 0.70 0.034 1.2 11.7 0.84 0.07
1.32
0.033 0.53
0.50
3.0
0.84 0.13 200 140 100 70
00
OCDM
| 0.33 0.65 0.045 1.3 12.0 0.80 0.17
1.39
|
0.48
0.52
1.8
0.89 0.13 200+180
100+9070
140
+140
|
1.0 0.65 0.045 1.0 10.0 0.67 0.5
1.16
|
1.0
1.0
2.0
1.57 0.15 370+340
SCDM
260 150 100
Table I: A set of representative quintessen e models satisfying all on ordan e onstraints. For models #2 5, w is the ee tive equation-of-state. The age t is in Gyr, while H0 t0 gives the age
in units of the Hubble time, whi h is equivalently the instantaneous power at whi h the s ale fa tor grows with time. The luminosity distan e in units of the0 Hubble length is given
by H0 d at
red shift z = 1; this quantity is dire tly related to the distan e modulus for supernovae. For omparison with the open, empty referen e osmology used in Figure 3, H0 d = 1:5. The magnitude
dieren e is then
(m M ) = 5 log10 (H0 d =1:5). The rate of hange of the equation-of-state for the tra ker quintessen e models, dw=dz,3+is evaluated at z = 0. The
growth fa tor is dened
as g  =
and the growth rate is f  d ln =d ln a. The normalization of the mass power spe trum is given by , where P (k) = 2 H k T 2 (k)=(4) and T is the baryoni and dark
matter transfer fun tion. The rms mass u tuation ex ess, 8 , 50 , 0is6evaluated for top hat window fun tions with radii 8; 50 Mp /h. For the0ve representative models, 8 agrees with the luster
abundan e onstraint given in Eq. (5). Also, the values of 8  8
as dened by Chiu et al. 1998 agree with their observational bound on  based on luster abundan es and pe uliar velo ities.
The rms bulk velo ity, in km/s, is evaluated for top hat radii
of R = 50; 150 Mp /h, with 2 Maxwellian error bars. We have not listed any quintessen e models lo ated in the region ex ised
between Figures 13 and 17 sin e models in this region are very similar to models 1 & 2. For omparison, the last two entries show the model parameters and properties of an open CDM model,
and SCDM. The tilt of the open model is not based on a parti ular open in ationary model but has been hosen so that the model is both luster and COBE (following Bunn & White 1997)
normalized.

other QCDM

3
4
5

s dm

:
m

ns

ns

COSMIC CONCORDANCE AND QUINTESSENCE

model
#
reeper QCDM / CDM 1

33